If I have factorial ANOVA related data but the data does not have a normal distribution, and also the transformation can not help me, what other analysis I can do with my data?
Why do you think that your "data" are non-normal? Note that not the *data* themselves (containing the effect), but the *residuals* after the ANOVA need to be normal.
If you, however, know that the process generating the data is the reason for non-normality or if you already know that the residuals are non-normal, you may consider to use a GLM (generalized linear model) or a resampling test.
Thomas
Thomas Petzoldt, thank you very much.
I test the residual as well, but not the data themselves nor the residuals have normal distribution! also I check the data for outliers and I remove them but the data didn't got the normality!
I am not familiar with generalized linear model, but the data do not have random effect! the class variables (fcators) have fix effect.
As said, the original data don't need to be normally distributed, only the residuals. In addition, removing outliers can be even worse, and you don't need to include random effects for using a GLM. I'm sorry, but I see a fundamental problem and don't think that this can be solved "remotely" on ResearchGate. Thomas
Yu Lin, thank you very much for your response,
but Kruskal-Valis is just for One-ANOVA in a free distribution manner while our experiment is three-way-ANOVA! for Two-Way-ANOVA we can use Two-Way-Nonparametric test but there is no such non-parametric test for three-way-ANOVA!
Do you think that I can perform Three-Way-ANOVA on ranking data? I mean data be transfromed to their rank and then perform ANOVA on ranked data.
It sounds like you are doing a designed experiment.
How many levels do you have for each of your factors?
What is the response?
If you have a continuous value, use MLR. If you have count values or concentrations, use Poisson Regression. If you have a % Yield, you should use a logit transform.
thank you every one, your comments are very useful.
Yes I have this problem with count as well as continuous data.
my main data is antioxidant activity of my plant but, I have count data such as tiller number as well.
My experiment is two-way factorial based on RB design (Three-Way-ANOVA).
if you are familira with SAS, please give me a proc to perform.
Are you allowed to post your data so we can see it?
If your response data is a count variable, then you should use a poisson regression or something similar to it. If you have continuous responses, then you could use multiple linear regression, survival analysis or a couple other things too.
Thank you Andrew.
I attached the data here with code for it analysis in SAS
Armin,
Are A, B and C the factor combinations? Are X1 to X5 the response variable?
yes ..... the a b and c are the the class factors and the R is the repeat of the data.
x1 - x5 are the respons variables.....
thtank you....
Hi Armin,
I was looking at some of your data sets. It looks like you have a couple outliers. That is normal. SInce you have about 100 samples, if the data is normally distributed, And we expect 3-7 data points beyond the 95% CI and 0-2 outside the 99% CI. I wouldn't worry too much.
- Badges
- Science topic